Author: Robert Hyatt
Date: 15:04:41 09/28/00
Go up one level in this thread
On September 28, 2000 at 10:34:29, Vincent Lejeune wrote: >On September 28, 2000 at 10:07:21, Robert Hyatt wrote: > >>On September 28, 2000 at 07:15:45, Vincent Lejeune wrote: >> >>>I would suggest a new type of "swindle mode" : the method is based on 1) make >>>the right move difficult to find for a human and 2) play quickly to put pressure >>>on human >>> >>>Here's the Idea : if an endgame is a tablebase draw, select all the drawing >>>moves and explore them only in tablebase to measure where the human will have >>>the hardest work to do : calculate, for all this drawing moves, with iterative >>>deepening, the worst ratio (number of drawing moves for human divide by number >>>of possible moves for human) for the human in the tree, because a narrower path >>>is difficult to follow (see the game Deep Thought-Alex Fishbein post some days >>>ago: where human had only one drawing move he didn't find it). The depth still >>>to be dertimined may be 5 plies, may be 10 or simply play right after the >>>human's move to put maximum pressure. >> >> >>I tried that a long way back. And it has a fatal flaw. Just because one side >>has only one move that holds the draw doesn't mean that is a difficult move to >>find. You would really want to find positions where the drawing move is very >>obscure or hard to understand, and that would be very hard to program. >> > >I found 2 flaws in current method : >1) it's not because a program consider a move as 'difficult' that a human do the >same >2) It's time consuming and time is very important in chess ... > >and it remains a strange thing : all the nodes analysed by Crafty are in TB ! > This last point is correct. I wrote a special search function, that was only used when in a drawn endgame position. The score backed up to the root was a minimaxed measure of how 'narrow' the drawing line was for the opponent. IE if at each ply the opponent has several drawing moves, the score would be lower than that for a line where the opponent had to play very accurately. The problem was that in many cases, just because there was only one move that held the draw, didn't mean it was hard to find. In fact, it was often very obvious. I gave up and chose to use the approach I use now... search all the draw moves without using tablebases... > >> >>> >>>This method exploit the main point of tablebase eg : the precision allied with >>>the speed; where the human brain get kill by machine. >>> >>>I hope that Bob will test this algorithm for Crafty , I'm confident in it :) >>> >>>If I was not enough clear please tell me... >>> >>>Best regards >>>Vincent Lejeune
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.